Can AI answer the $3T revenue question required to pay for the chips?
sequoia's david cahn estimates the ai industry needs $3t in revenue to justify projected 2026 infrastructure spending. falling token prices and cheaper models make this payoff increasingly difficult.
sequoia's david cahn updated his analysis on ai infrastructure spend, projecting that the industry must earn $3 trillion to cover the $1.5 trillion capex for chips and data centers expected by 2026. this required revenue per gigawatt has risen due to memory costs and bottlenecks, with current leaders like openai and anthropic falling short of the necessary volume. apollo chief economist torsten slok highlighted risks to this payback, noting that hyperscalers predict cash flow accelerations by 2028 but face headwinds from cheaper open-weight models and rapidly declining token prices. slok warns that failure to meet these revenue targets could tip the economy into recession and force an s&p 500 correction, as the entire market rally currently depends on these few companies.